CHEMICALS

Fouling predictions that carry their own confidence.

Fouling models either drift or cannot see past their training data, and both give one number with no error bar. EntroMetrix gives every forecast a confidence range and finds a cleaning plan you can defend.

DOWNLOAD PDF REPORT All case studies

KEY RESULTS

−41%

prediction error against a model built on data alone

−58%

training data needed to reach the target accuracy

−34%

narrower confidence ranges after calibration

−64%

fewer model runs to find the best cleaning plan

+38%

longer warning before the train runs inefficiently

01 · THE PROBLEM

A number with no error bar is hard to act on.

Models built on physics are faithful, but their fouling parameters drift as crude and duty change. Models built on data alone need long, clean histories and cannot see beyond the conditions they were trained on.

Both give one number with no sense of how far to trust it. Yet the decisions that rest on them, when to take a shell out to clean and how hard to fire the furnace, are costly and slow to reverse.

01The twin tracks the measured temperature where a data only model drifts, and finds the best plan in few runs.

02 · WHAT IT COSTS YOU TODAY

Cleans called too early, or too late.

Fuel and carbon: as the train fouls, the furnace fires harder to make up the lost preheat.

Throughput: when the furnace reaches its limit, crude rate is cut until a clean.

Cleaning: a shell pulled on a guess costs the clean and the duty lost while it is out, and a shell left too long costs more than the clean.

02Forecast coil inlet temperature with its confidence band.

03 · WHAT ENTROMETRIX DOES

Physics at the core, calibrated to your history.

Builds the twin: a mechanistic law of deposit build up and removal, with a small learned correction for what the physics misses, so it stays accurate on little data.

Calibrates it: the physical parameters are inferred from your own history as ranges, so every forecast carries a confidence range.

Values every clean: for each shell, the temperature a clean recovers, the fuel and carbon it saves and the duty lost while it is out, each as a range.

Plans the whole train: cleaning one shell changes the duty on the others, so the plan weighs fuel, carbon, throughput and cleaning cost together and returns a set of plans to choose from.

03Fuel, carbon, cleaning and total cost against current practice.

04 · HOW YOU STAY IN CONTROL

Checked before it informs a decision.

Your history: calibrated on part of your record and tested on runs it did not see.

Your parameters: deposit, removal and ageing constants your engineers can read and check.

Your confidence check: the stated confidence is tested against measured operation, so it means what it says.

Your systems: runs on historical data beside the plant, never in the control loop.

Your decisions: the plan is a recommendation, and your engineers decide which shell to clean and when.

04The value of cleaning each shell, with its range.

05 · WHAT YOU GAIN

Predictions you can act on.

Accurate Where Data Is Thin

Holds its accuracy with less than a fifth of the record, where a data only model climbs steeply.

A Range, Not a Guess

Every forecast comes with a calibrated range, so you know how far to trust it.

A Plan for the Whole Train

Priority cleans with their windows, and the rest left to the turnaround.

05Which shell to clean and when, to the next turnaround.

DOWNLOAD PDF REPORT All case studies

CASE STUDIES

CHEMICALS

See what the model finds in your plant.

A short call with our engineering team, your process data stays on site.

Request a demo See the science

A number with no error bar is hard to act on.

Cleans called too early, or too late.

Physics at the core, calibrated to your history.

Checked before it informs a decision.

Predictions you can act on.

Energy and carbon optimisation on the crude distillation unit.

Real time optimisation of the cat cracker.

Disruption seen before the line stops.

See what the model finds in your plant.